QoS-Oriented Grid-Enabled Data Warehouses - Data Warehousing Design and Advanced Engineering Applications - page 164

Figure 6. Experimentally tested network topologies

in the most powerful one. Figure 5 presents the
number of queries submitted at each site. In order
to model that, we consider that each query may be
submitted by a user from any site, but the prob-
ability of a query being submitted by a specific
site is proportional to the number of DW users
at the site. Users' access patterns were modeled
considering that half of the tasks access data stored
at the same site that submitted the job.
In order to evaluate the effectiveness of the
QoS-aware scheduling and dynamic replication
strategies, we have made several tests using the
QoS-oriented scheduling and three different data
dynamic replication strategies: (i) the QoS-aware,
(ii) Best Client (BC) and (iii) Data Least Loaded
(DLL). Variations of the BC strategy are used by
Ranganathan & Foster (2001) and Siva Sathya
et al (2006). The DLL is used by Ranganathan &
Foster (2004). Both in the BC and in DLL, each
site monitors the number of data access at the data
replicas it stores, computing the number of times
the fragment was requested. When such number is
greater than a threshold value, a fragment replica
is created at another site. In BC, the fragment
replica is created at the site that has more times
demanded for the considered fragment. In DLL,
fragment replica is created at the site that has the
least remaining work to execute.
The obtained SLO-AR values are represented
in Figure 7. All the evaluated dynamic replication

methods lead to good SLO-AR. Such success
ensures the quality of the used QoS-oriented
hierarchical scheduling model. But the obtained
results also present the benefits of the proposed
QoS-aware dynamic data replication and place-
ment, as it was the method that leads to the highest
value of SLO-AR.
In Figure 8 we present the measured through-
put for the evaluated strategies. Once again, the
proposed QoS-aware replication strategy leaded
to the highest values in the two tested network
configurations. This happens because such place-
ment strategy has the same objective of the used
query scheduling strategy. Therefore, not only it
increases the number of queries that the Commu-
nity Scheduler agrees to execute but also lead to
a better resource utilization than the other replica
selection and placement methods.
The number of created replicas per method was
almost the same and no method created a huge
number of replicas for the same data fragment
(as show in Figure 9). In fact, the success of the
QoS-aware scheduling method is mostly related
to its replica placement strategy.
In Figure 10, we present the number of created
replicas per site for the Hierarchical Topology.
Let's analyze site 6: such site is almost the most
powerful one and it is the site that have the high-
est number of submitted queries. This is also the
site on which the BC method created the highest